A grey relational analysis based evaluation metric for image captioning and video captioning

摘要

Aiming at the performance evaluation on image captioning and video captioning, this paper discusses the existing performance metrics and then suggests a novel overall performance metric based on grey relational analysis of Grey System Theory. In our metric, all the available performance metrics of each captioning model is used to extract a comparative sequence. Meanwhile, a reference sequence is constructed by combining all the available performance metrics involved. Then an overall metric is obtained by computing the grey relational degrees between the two kinds of sequences. Experimental results on the most widely-used image dataset and video dataset show that the proposed metric is fast and effective that helps for spurring improvements and measuring progress in the state of the art of image captioning and video captioning.